GIFT-Eval
收藏资源简介:
GIFT-Eval是由Salesforce AI Research创建的一个通用时间序列预测模型评估基准。该数据集包含144,000条时间序列,涵盖七个领域、10种频率,数据点总数达1.77亿。数据集的创建过程包括了广泛的预训练和训练/测试组件,确保了多样性和广泛的应用场景。GIFT-Eval旨在解决时间序列预测模型在零样本和通用预测能力评估中的不足,推动时间序列基础模型的发展。
GIFT-Eval is a general-purpose time series forecasting model evaluation benchmark created by Salesforce AI Research. This dataset contains 144,000 time series spanning seven domains and 10 frequencies, with a total of 177 million data points. The construction of this dataset incorporates extensive pre-training workflows and properly structured train/test components to ensure its diversity and broad coverage of application scenarios. GIFT-Eval aims to address the gaps in evaluating the zero-shot and general forecasting capabilities of time series forecasting models, and promote the development of time series foundation models.




